All work
Fraimed
Agents lose history. Decisions get reopened. Fraimed is the memory layer I built so the work that already happened is still available to the work that comes next.
Why I built it
It came out of a problem I kept hitting in my own agentic work. Agents lose context. Decisions disappear into old conversations. The same ground gets covered again because nothing carried forward.
Fraimed is continuity. What happened, what was decided, and what should still be known about it later. Not a transcript, and not everything that was ever said, but the part worth treating as true tomorrow.
How it works
The sequence is the same every time, and it is short enough to describe rather than illustrate stage by stage.
- An agent starts cold
- A task begins with incomplete information. The decision that settled this three weeks ago sits in a conversation nobody is going to reopen, so the work either stalls or gets redone against different assumptions.
- Fraimed recovers the history
- Relevant prior history is retrieved for the work at hand rather than dumped in wholesale, with a reference back to where it came from.
- New work produces something worth keeping
- Decisions and insights from the work can be promoted into durable memory. Sophie is where that curation happens. Fraimed is where accepted memory lands.
- Promotion stays a decision
- Not everything said deserves to become something an agent treats as true tomorrow, so a person accepts, edits, or rejects each candidate.
Where Fraimed sits
AgentSophieFraimedThe work
Decisions that matter
- Retrieve what matters, not everything
- Recovering all prior context is the same problem as having none. Retrieval is scoped to the work at hand.
- Every entry carries its source
- Knowledge an agent cannot trace is knowledge nobody can correct.
- Promotion is a decision
- Nothing becomes durable because it was said. It becomes durable because it was accepted.
- Useful without Sophie
- Sophie is the interaction layer I built first, but Fraimed is its own product.